Code for KHGT model, AAAI2021

Last update: Nov 29, 2022

Related tags

Recommender Systems KHGT

Overview

KHGT

Code for KHGT accepted by AAAI2021

Please unzip the data files in Datasets/ first.

To run KHGT on Yelp data, use

python labcode_yelp.py

For MovieLens data, use the following command to train

python labcode_ml10m.py --data ml10m --graphSampleN 1000 --save_path XXX

and use this command to test with larger sampled sub-graphs

python labcode_ml10m.py --data ml10m --graphSampleN 5000 --epoch 0 --load_model XXX

For Online Retail data, use this command to train

python labcode_retail.py --data retail --graphSampleN 15000 --reg 1e-1 --save_path XXX

and also load the model to test it with larger sampled sub-graphs

python labcode_retail.py --data retail --graphSampleN 30000 --epoch 0 --load_model XXX

Comments

Issue run retail dataset
Hi, Thank you for your quick response. I was unable to run the code on retail dataset. (and Successfully run the code on ml10m and yelp datasets ) BTW, I have fixed the following minor issues to enable the correctness:

change the folder name from 'Yelp' to 'yelp' (match the folder name in DataHandler_time.py)

change line 14 in DataHandler_time.py to 'retail' (instead of 'Tmall')

During the execution of

python labcode_retail.py --data retail --graphSampleN 15000 --reg 1e-1 --save_path model_name

The error pops out due to the index out of range. I think these are some issues with the dataset. Please let me know if you could get it run. And would you mind to sharing me the original data and data preprocessing code?
opened by KylinA1 3

The axis of softmax supposed to be 0?

def GAT(self, srcEmbeds, tgtEmbeds, tgtNodes, maxNum, Qs, Ks, Vs):
	QWeight = tf.nn.softmax(NNs.defineRandomNameParam([args.memosize, 1, 1], reg=True), axis=1)
	KWeight = tf.nn.softmax(NNs.defineRandomNameParam([args.memosize, 1, 1], reg=True), axis=1)
	VWeight = tf.nn.softmax(NNs.defineRandomNameParam([args.memosize, 1, 1], reg=True), axis=1)
	Q = tf.reduce_sum(Qs * QWeight, axis=0)
	K = tf.reduce_sum(Ks * KWeight, axis=0)
	V = tf.reduce_sum(Vs * VWeight, axis=0)

I found above code to implement equation (3), I have 2 questions:

I think it should be below:

tf.nn.softmax(..., axis = 0)

For all behavior $k$, there supposed to be only 1 group of $Q^h_m, m = 1\cdots, M$ to aggregated from, according to equation (3). While in practical implementation, each behavior $k$ has its own $M$ channels basis parameters.

Please excuse me if there is any misunderstanding.

opened by KylinA1 1

Datasets problem

Dear author, I would like to ask you some problems in datasets that is, the user of each dataset exceed 10000, but I found that only 10000 users in test data For example, there are 67788 users in ML10M, but only 10000 users are chosen for test, is it reasonable? Thanks!

opened by Weile0409 1
embedding of uEmbed0/iEmbed0 and UEmbedPred/IEmbedPred

Hello I have a question about embedding of uEmbed0/iEmbed0 and UEmbedPred/IEmbedPred.

In "ours" function, why Embed0 is replaced by a random initialized EmbedPred in code line 125~126 in labcode_ml10m.py？I didn't find a description of it in the paper, and I don't understand what the motivation is for doing so.

I am confused.

Anyway, thanks of the codes. It is pretty.

opened by fsgdrq 1
embedding of srcNodes and tgtNodes

Hello, I have a question. In the "messagePropagate" function, why do srcNodes and tgtNodes look up their embeddings in the same embedding table?

opened by GhostShipZ 1

Code for KHGT model, AAAI2021

Related tags

Overview

KHGT

Comments

Issue run retail dataset

The axis of softmax supposed to be 0?

Datasets problem

embedding of uEmbed0/iEmbed0 and UEmbedPred/IEmbedPred

embedding of srcNodes and tgtNodes

Owner

[AAAI2021] The source code for our paper 《Enhancing Unsupervised Video Representation Learning by Decoupling the Scene and the Motion》.

Implementation of our paper 'RESA: Recurrent Feature-Shift Aggregator for Lane Detection' in AAAI2021.

Official implementation of "Dynamic Anchor Learning for Arbitrary-Oriented Object Detection" (AAAI2021).

Simple is not Easy: A Simple Strong Baseline for TextVQA and TextCaps[AAAI2021]

Implementation for our AAAI2021 paper (Entity Structure Within and Throughout: Modeling Mention Dependencies for Document-Level Relation Extraction).

Out-of-Town Recommendation with Travel Intention Modeling (AAAI2021)

This is the official source code for SLATE. We provide the code for the model, the training code, and a dataset loader for the 3D Shapes dataset. This code is implemented in Pytorch.

In this project we investigate the performance of the SetCon model on realistic video footage. Therefore, we implemented the model in PyTorch and tested the model on two example videos.

Step by Step on how to create an vision recognition model using LOBE.ai, export the model and run the model in an Azure Function

Capture all information throughout your model's development in a reproducible way and tie results directly to the model code!

This is discord nitro code generator and checker made with python. This will generate nitro codes and checks if the code is valid or not. If code is valid then it will print the code leaving 2 lines and if not then it will print '*'.

Inference code for "StylePeople: A Generative Model of Fullbody Human Avatars" paper. This code is for the part of the paper describing video-based avatars.

This is the code for our KILT leaderboard submission to the T-REx and zsRE tasks. It includes code for training a DPR model then continuing training with RAG.

Reference implementation of code generation projects from Facebook AI Research. General toolkit to apply machine learning to code, from dataset creation to model training and evaluation. Comes with pretrained models.

Code for CodeT5: a new code-aware pre-trained encoder-decoder model.

Model search is a framework that implements AutoML algorithms for model architecture search at scale

Model summary in PyTorch similar to `model.summary()` in Keras

Implementation of STAM (Space Time Attention Model), a pure and simple attention model that reaches SOTA for video classification

ReConsider is a re-ranking model that re-ranks the top-K (passage, answer-span) predictions of an Open-Domain QA Model like DPR (Karpukhin et al., 2020).